CCenergy DIIS manipulation

Hello friends!,
I’m dealing with a highly multireference diatomic electronic state (because that’s what I do) and having some CCSD convergence issues. The ^4Pi state of FeH seems to behave with some basis sets and not with others. Using cc-pwCVQZ, I get a good geometry and correlation energy/relative energy to other electronic states that matches what I have observed in molpro. At cc-pwCVTZ and cc-pwCV5Z the CCSD energy is 0.2 Hartrees to high. I have looked at the MO coefficients of the ROHF wave function - it looks like the correct state and ROHF energy, and a stability analysis doesn’t find a better set of orbitals. I guess the problem is with the CC convergence. I tried turning off DIIS and ccenergy went kray.

Is it possible in PSI4 to delay when DIIS works, and/or play with the number of expansion vectors? This does a great job of solving fussy CC convergence problems in molpro and cfour.


PS: input file

molecule {
0 4
H 1 1.562441

set {
reference rohf
docc [7, 0, 3, 2]
socc [1, 1, 0, 1]
scf_type pk
print_mos true
rms_force_g_convergence 2
freeze_core false

assign Fe cc-pwcvtz
assign H cc-pwcvtz

set scf df_scf_guess false
set scf guess gwh
set ccenergy r_convergence 10

Right now the DIIS algorithm in the CC codes is hardwired to 8 error vectors and immediate extrapolation. However, this would be trivially easy to modify with a couple of new input parameters.